diff --git a/dlib/dnn/trainer.h b/dlib/dnn/trainer.h
index 337099d73..84b311ce3 100644
--- a/dlib/dnn/trainer.h
+++ b/dlib/dnn/trainer.h
@@ -131,9 +131,10 @@ namespace dlib
         }
 
         net_type& get_net (
-        ) const 
+        )  
         { 
             wait_for_thread_to_pause();
+            sync_to_disk(true);
             propagate_exception();
             return net; 
         }
@@ -266,7 +267,6 @@ namespace dlib
         {
             DLIB_CASSERT(data.size() == labels.size() && data.size() > 0);
 
-            bool updated_the_network = false;
             // The reason these two loops don't initialize their counter variables but
             // instead use class members is so we can include the state of the loops in the
             // stuff written by sync_to_disk()
@@ -297,7 +297,6 @@ namespace dlib
                     send_job(data.begin()+epoch_pos, 
                               data.begin()+std::min(epoch_pos+mini_batch_size,data.size()), 
                               labels.begin()+epoch_pos);
-                    updated_the_network = true;
                 }
                 epoch_pos = 0;
 
@@ -313,7 +312,7 @@ namespace dlib
             }
             wait_for_thread_to_pause();
             // if we modified the network at all then be sure to sync the final result.
-            sync_to_disk(updated_the_network);
+            sync_to_disk(true);
         }
 
         void train (
@@ -326,7 +325,6 @@ namespace dlib
             static_assert(has_unsupervised_loss, 
                 "You can only call this version of train() when using an unsupervised loss.");
 
-            bool updated_the_network = false;
             // The reason these two loops don't initialize their counter variables but
             // instead use class members is so we can include the state of the loops in the
             // stuff written by sync_to_disk()
@@ -356,7 +354,6 @@ namespace dlib
                     sync_to_disk();
                     send_job(data.begin()+epoch_pos, 
                              data.begin()+std::min(epoch_pos+mini_batch_size,data.size()));
-                    updated_the_network = true;
                 }
                 epoch_pos = 0;
 
@@ -372,7 +369,7 @@ namespace dlib
             }
             wait_for_thread_to_pause();
             // if we modified the network at all then be sure to sync the final result.
-            sync_to_disk(updated_the_network);
+            sync_to_disk(true);
         }
 
         void set_synchronization_file (
@@ -588,6 +585,7 @@ namespace dlib
             main_iteration_counter = 0;
             while(job_pipe.dequeue(next_job))
             {
+                updated_net_since_last_sync = true;
                 ++main_iteration_counter;
                 // Call compute_parameter_gradients() and update_parameters() but pick the
                 // right version for unsupervised or supervised training based on the type
@@ -746,6 +744,7 @@ namespace dlib
             prob_loss_increasing_thresh_default_value = 0.99;
             prob_loss_increasing_thresh_max_value = 0.99999;
             prob_loss_increasing_thresh = prob_loss_increasing_thresh_default_value;
+            updated_net_since_last_sync = false;
             start();
         }
 
@@ -832,10 +831,15 @@ namespace dlib
                 dlib::cuda::set_device(prev_dev);
             }
         }
+
         void sync_to_disk (
             bool do_it_now = false
-        )
+        ) 
         {
+            // don't sync anything if we haven't updated the network since the last sync
+            if (!updated_net_since_last_sync)
+                return;
+
             // If the sync file isn't set then don't do anything.
             if (sync_filename.size() == 0)
                 return;
@@ -879,6 +883,7 @@ namespace dlib
 
                 last_sync_time = std::chrono::system_clock::now();
                 main_iteration_counter_at_last_disk_sync = main_iteration_counter;
+                updated_net_since_last_sync = false;
             }
         }
 
@@ -1069,6 +1074,7 @@ namespace dlib
         double prob_loss_increasing_thresh_default_value;
         double prob_loss_increasing_thresh_max_value;
         double prob_loss_increasing_thresh;
+        std::atomic<bool> updated_net_since_last_sync;
 
     };
 
diff --git a/dlib/dnn/trainer_abstract.h b/dlib/dnn/trainer_abstract.h
index 1faf9548e..24c25fb1d 100644
--- a/dlib/dnn/trainer_abstract.h
+++ b/dlib/dnn/trainer_abstract.h
@@ -88,8 +88,8 @@ namespace dlib
                       cuda_extra_devices.
         !*/
 
-       net_type& get_net (
-        ) const; 
+        net_type& get_net (
+        ); 
         /*!
             ensures
                 - returns the neural network object used by this trainer.  This is the
@@ -99,6 +99,8 @@ namespace dlib
                   dnn_trainer's constructor.
                 - This function blocks until all threads inside the dnn_trainer have
                   stopped touching the net. 
+                - This function will sync the trainer state to disk if the current state 
+                  hasn't already been synced to disk since the last network modification.
         !*/
 
         const std::vector<solver_type>& get_solvers (